Intel Speed Binning

[Dopekitten]

Ok, so I'm curious, Intel obviously tests all their chips--both to make sure they work and to speed bin them--and I know that.

What I don't know is how they can possibly be efficient in testing that many CPU's. I mean, how do they get the CPU's under a heatsink/in a motherboard fast enough?

Do they have hundreds of sweatworkers screwing on heatsinks all day?

 

[JAG87]

...

the whole fab is full of robotic machines and you think they don't have a machine that grabs each chip, presses it on a socket, runs a quick test and determines voltage frequency and tdp?

they probably have a machine that counts all the hairs on your butt in 5 seconds.


here I found a video for you:
http://www.youtube.com/watch?v=NVhxmQKQbVI

 

[JFAMD]

I can't speak to their fabs, but in the industry, test and marking happens just after packaging.

Chips are run through testers, they are marked and fused at top speed that they pass at, unless there is a need for lower speeds (higher demand). It is all automated.

 

[dmens]

...

here I found a video for you:
http://www.youtube.com/watch?v=NVhxmQKQbVI

LOL I work for the 3rd guy in the video.

 

[Nemesis 1]

AsI recall it Intel changed how they ben with the merom product , As I recall intel changed there top been parts to have the best efficiency

 

[SanDiegoPC]

LOL I work for the 3rd guy in the video.

You work for INtel?

That was an awesome video!! totally robotic as you'd expect - and with a testing software that has 2M lines of code!!

Question: Three times in that vid I think they mention "video processing" capability of the Westmere 32nm chip. Are they trying to integrate graphics onboard the CPU so we don't need a seperate video card?

 

[Phynaz]

You work for INtel?

That was an awesome video!! totally robotic as you'd expect - and with a testing software that has 2M lines of code!!

Question: Three times in that vid I think they mention "video processing" capability of the Westmere 32nm chip. Are they trying to integrate graphics onboard the CPU so we don't need a seperate video card?

Not trying, it's done. At least it's on the same package as the chip, not in the same die, yet.

 

[Markfw]

LOL I work for the 3rd guy in the video.

If I counted right, is that Dave Obrien ?

 

[cbn]

...

the whole fab is full of robotic machines and you think they don't have a machine that grabs each chip, presses it on a socket, runs a quick test and determines voltage frequency and tdp?

they probably have a machine that counts all the hairs on your butt in 5 seconds.


here I found a video for you:
http://www.youtube.com/watch?v=NVhxmQKQbVI

Thank you for the video.

Since that binning process is very sophisticated does this guarantee a Core i7 940/950/960 CPU will always overclock better than a 920 chip (provided the stepping is the same)?

 

[nutxo]

Thank you for the video.

Since that binning process is very sophisticated does this guarantee a Core i7 940/950/960 CPU will always overclock better than a 920 chip (provided the stepping is the same)?

Not at all. Quite often the lower binned cpus will OC equal to what the higher will OC to for some reason. Ive always found the highest binned cpus to be mediocre overclockers. I don't know why.

 

[veri745]

Not at all. Quite often the lower binned cpus will OC equal to what the higher will OC to for some reason. Ive always found the highest binned cpus to be mediocre overclockers. I don't know why.

Much of the time cpus are limited by power, not speed. Many parts are binned lower because they can't make power, but they may be way faster (especially if you up the voltage).

 

[dmens]

If I counted right, is that Dave Obrien ?

Not him, but I know the guy. I work in the tester debug lab.

 

[JFAMD]

Not at all. Quite often the lower binned cpus will OC equal to what the higher will OC to for some reason. Ive always found the highest binned cpus to be mediocre overclockers. I don't know why.

Often, if you are yeilding higher than normal, you end up down binning. Let's say your normal distribution was 15% on top bin. Then you do some process improvements, and you get to 35% top bin. And maybe N-1 is up to 85%, and by N-2 you are at 100% yield. Well, every chip coming off the line is capable of at least N-2 speed at that point, but you need a bunch of N-3 and N-4 because your demand is coming in for those speeds. You end up down binning the N-2 material to N-3 or N-4. They overclock well because they could easily be rated at N-2 or better.

The higher the bin speed, the more you are "pushing the envelope" so there is probably less headroom. This is all from a proportional standing of course.

Processors are like snowflakes, so don't expect that you can just buy the low end stuff and overclock like crazy because sometimes an N-7 is never gonna get above N-7.

 

[pm]

Beyond just down-binning - which in my experience isn't all that common - overclocking takes advantage of the fact that CPU's are tested for reliability for use in low-end systems used in higher temperatures and systems with less-than-perfect power-supplies. There's also guardband - which basically is the principal that you don't test the part to it's absolute limits. Overclockers also tend to increase the voltage above reference and that reduces long-term reliability and increases power - both of which the manufacturer avoids.

Not him, but I know the guy. I work in the tester debug lab.

One these days, we are likely to meet. I've been calling into some of your team's meetings.

 

[jvroig]

If I counted right, is that Dave Obrien ?

Third guy in the vid is Mick Hancock of the debug lab.

@JFAMD : Nice post. Clear and concise, gets the point across rather well. Also, did you watch the vid? Would you know if AMD's testing procedures vary or is it identical?

 

[JFAMD]

Didn't watch the video. Won't really comment publicly on any testing methodology beyond what I have said here. I have been to our test facilities in Singapore and have seen our testers in action, but don't work in that side of the business.